Lambert calls the license 'trash' and objects to branding and usage restrictions. He calls one clause reasonable and qualifies a concern about output-trained models. The label describes his overall criticism of the licensing terms.
Role at the time: AI researcher and Interconnects author
Challenged the characterization
Statement summary
On September 19, Nathan Lambert said that near-term inference-time-compute scaling should be distinguished from recursive self-improvement and that he had not seen enough evidence for a near-term intelligence explosion.
“I was underestimating how much we are likely to scale inference-time compute in the near future, but this is different than RSI and I have not seen enough to convince me that an intelligence explosion is near.”
Lambert disputes the recursive-self-improvement and near-term-intelligence-explosion characterization without defending or condemning Anthropic’s conduct.
Role at the time: Personal public post by Nathan Lambert; not established as an institutional statement.
Challenged the characterization
Statement summary
Lambert challenges the evidentiary basis for an extinction probability near 10%, while acknowledging serious AI safety challenges. He argues that current evidence about AI improving training software and performing superhumanly in some domains does not establish near-term human extinction.
“but am still waiting for evidence that supports anything close to a 10% chance of human extinction.”
Meta promotes broad access to Muse Spark Open. Lambert criticizes its licensing restrictions while accepting the revenue gate and qualifying one concern about downstream models.
After leaving Anthropic, Jacob Coxon said Anthropic and OpenAI were irresponsibly racing toward self-improving superintelligence, while praising Anthropic's current practices. Responses dispute catastrophic-risk forecasts, working inside AI labs, and whether governments should pause or regulate development.
